Because of the characteristic vulnerability of persons with dementia, families often play a key role in their lives in nursing facilities. Families are generally an underutilized resource, and insufficient attention is given to the family 's role in improving the quality of life of residents in SCUs. The goal of this study is to evaluate the outcomes on residents and family that result initiation of a formal partnership between families and staff in SCUs. This Family Partnership Program has been designed by a multidisciplinary team with extensive experience in the care of patients with Alzheimer's Disease, interactions with families, and the organization management of SCUs. The program will be refined and pilot-tested at HRCA prior to implementation. To test the effect of the Family Partnership Program a controlled trial will be conducted in 18 nursing home-based SCUs in Massachusetts, New Hampshire and Rhode Island. The sample will consist of 234 Experimentals and their two most involved family members, contrasted with 234 Controls. Resident selection criteria include: presence of functional deficit, cognitive status, presence of a family member, and expected stay in the SCU. Selected residents be assessed at four points in time: baseline, 5 months, 13 months, and 19 months. Residents in SCUs that have been randomly designated as Experimentals will begin exposure to the Family Partnership Program following the month 5 assessment. Among resident outcomes are: change in cognitive status, mood status, involvement patterns, social response patterns, physical functioning, and health complications. For family members, outcome areas include: satisfaction with care, level of involvement, nature of involvement and secondary morbidity. We have used a carefully designed research approach to ensure our ability to measure the impact of the Family Partnership Program intervention. This includes a stratified sample with matched Experimental/Control pairing of Special Care Units, randomization, covariate adjustment, multiple repeated measures, and profile comparisons within a MANCOVA model. Our methods should be sufficient to permit the generation of comparisons that are not compromised because of the threat of selection bias, regression to the mean, and other major problems of internal validity.